CN103279769B - A kind of characteristics of objects expression of doing more physical exercises being applicable to different scene - Google Patents
A kind of characteristics of objects expression of doing more physical exercises being applicable to different scene Download PDFInfo
- Publication number
- CN103279769B CN103279769B CN201310232933.9A CN201310232933A CN103279769B CN 103279769 B CN103279769 B CN 103279769B CN 201310232933 A CN201310232933 A CN 201310232933A CN 103279769 B CN103279769 B CN 103279769B
- Authority
- CN
- China
- Prior art keywords
- bending moment
- value
- moving objects
- moment value
- scene
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of characteristics of objects expression of doing more physical exercises being applicable to different scene, for the feature that different Moving Objects features is different, propose self-adaptation combined invariant moment value method, dynamic select not bending moment value for describing the feature of different motion object.By defining same quefrency-inverse singular frequency method, being called for short SF-ISF method, calculating the weighted value of the not bending moment value of each object, afterwards again using the weighted value of not bending moment value and combined invariant moment value as input parameter; Set up multi classifier model, the multi-motion object in scene is classified.The present invention can reduce computing time effectively, high to the discrimination of Moving Objects, is applicable to identify the Moving Objects in monitoring in real time, can be applicable to multiple different video monitoring scene.
Description
Technical field
The invention belongs to technical field of image processing, be specifically related to a kind of recognition methods of Moving Objects.
Background technology
Square in statistics for characterizing the distribution of random quantity, if binary map or gray-scale map are regarded as two-dimentional density fonction, its characteristics of image square describes, moment characteristics belongs to one of provincial characteristics, bending moment theory is not by extracting the mathematical feature with translation, Invariant to rotation and scale of image, carries out image recognition.Bending moment is not theoretical proposed by 1962, be developed so far, continuous evolution and improvement, define very various type, and the not bending moment of every type has the speciality data corresponding to it to calculate category, also have different contributions corresponding to the different values of same class not bending moment to the calculating of data similarity.
Yu Jihong, Lv Junwei, " a kind of new ship images target identification method based on combined invariant moment " of Bai Xiaoming work, infrared, 2011,9 (32): 23-28 propose the partial quantity in Hu not bending moment, Wavelet Invariant Moment and affine not bending moment to combine Expressive Features, and the not bending moment array mode that employing is fixed describes the Moving Objects in its single scene, there is the not bending moment value of identical weighed combination, adopt simple Euclidean distance to carry out Classification and Identification Moving Objects, Zeng Wanmei, Wu Qingxian, " Air Target Identification based on combined invariant moment feature " of Jiang Changsheng work, electric light and control, 2009, 7 (16): 21-24, 44 propose affine for NMI characteristic sum not bending moment to combine, the not bending moment array mode that employing is fixed describes the Moving Objects in its single scene, there is the not bending moment value of identical weighed combination, Fisher is adopted linearly to judge the Moving Objects recognition methods of classifier methods, Liu Zhengjun, Li Qi, " the laser radar Range Profile target identification based on combination square " of Wang Qi work, Chinese laser, 2012, the not bending moment array mode that 6 (39): 1-7 employings are fixed describes the Moving Objects in its single scene, use the combination Moment Methods that BP neural net method training sample obtains in conjunction with Hu not bending moment and affine not bending moment.The sorting technique that dissimilar not bending moment combines all is aimed at specific Moving Objects in the special scenes of specific area by above method, and field of video monitoring widely, be applicable to land and water transportation, community, the scene that intelligent building etc. are different, need the Moving Objects of identification also of a great variety, in order to the different motion object of different scene of classifying, the invariant moment features value adopted is more, do not represent recognition capability stronger, redundancy may to be there is in the set of all not bending moment values, low contribution, not bending moment value useless even, these unnecessary not bending moment values can reduce the discrimination to Moving Objects.
Summary of the invention
The object of the present invention is to provide a kind of characteristics of objects expression of doing more physical exercises being applicable to different scene, improve the discrimination to Moving Objects.
In order to solve above technical matters, the technical solution used in the present invention is as follows:
Be applicable to a characteristics of objects expression of doing more physical exercises for different scene, it is characterized in that comprising the following steps:
The first step: the situation that the present invention discusses is very high for the compliance of Moving Objects not bending moment, not very sensitive to numerical accuracy, and different video may have different resolution, so do not need too complicated calculating and refinement problem, by all kinds of initial not bending moment value of object of doing more physical exercises in scene
,
represent i-th of m object initial not bending moment value, taking absolute value according to formula (1) to not bending moment and take the logarithm calculates not bending moment value
,
represent i-th of m object not bending moment value
(1)
Second step: the not bending moment value calculating each Moving Objects in whole Moving Objects set
same quefrency-inverse singular frequency
value
(2)
In formula
represent not bending moment value
the frequency occurred in same Moving Objects; N represents the sum of all Moving Objects in Moving Objects set;
represent the frequency occurring current not bending moment value in different motion object set, by each in Moving Objects set not bending moment value carry out above-mentioned analysis, obtain each not bending moment value of each Moving Objects
value; And then utilize
value sets up vector model respectively for each Moving Objects, as the criteria for classification of feature and the related check standard of each dimension;
3rd step: because vector model is that higher-dimension and extreme are sparse, according to information theory,
value be the cross entropy of not bending moment value probability distribution under specified conditions,
value be used to increase the weight of not bending moment value, with the data of expressive movement characteristics of objects better, therefore pass through
the filter operation of value, can select some important not bending moment values, carry out characterizing motility object with this from each Moving Objects; This makes it possible to accomplish under guarantee does not affect the prerequisite of Moving Objects feature extraction, reduce the dimension that Moving Objects proper vector represents most possibly; According to the vector model described in step 2, to not bending moment value
value sorts, and therefrom chooses the superseded threshold value being greater than user's setting
's
value; Select
the key not bending moment value that value is corresponding
,
represent a jth key not bending moment value of m object, 1≤j≤J, J is the number of all keys not bending moment value, will
as the feature of Moving Objects;
4th step: if the not bending moment value that in two Moving Objects, similarity is higher is each other more, and not shared by bending moment value
value ratio in respective Moving Objects is higher, illustrate these not bending moment value more can reflect their importance in Moving Objects, calculate a jth key not bending moment value of m Moving Objects of described step one Scene
weight sets
,
,
computing method as shown in formula (3), 1≤i≤m
(3)
Described
computing method as shown in formula (4)
(4)
In formula (4)
represent key not bending moment value
's
value;
for similarity exceedes the similarity threshold of user's setting
key not bending moment value weight
set
(5)
5th step: because bending moment value does not represent most important character representation in a Moving Objects, therefore the not bending moment value weight of scene just can by Moving Objects not bending moment value weight come out, calculate described step one Scene not bending moment vector weighted value
(6)
Calculate this scene according to formula (6) and choose all not bending moment values
6th step: by Weight
combined invariant moment value
put into the training of multi classifier model, obtain training pattern by training study, using the combined invariant moment value in practical application as input parameter, calculate in the model after input training, can obtain doing more physical exercises in scene object classification result.
Described frequency calculation method is as follows:
represent the not bending moment of m object,
represent i-th of m object not bending moment value, by all Moving Objects
value is mapped on linear coordinate, calculates the average of all not bending moment values
, not bending moment value
frequency be
the number of other the not bending moment values than itself in scope.
Described same quefrency is certain numerical value that bending moment value is not adjacent in same Moving Objects; Adjacent numerical value is more, and bending moment value is more not relevant with the character representation of this Moving Objects to represent this.
Described inverse singular frequency is certain numerical value that bending moment value is not adjacent in different Moving Objects coupling; Adjacent numerical value is more, represents that the separating capacity of this not bending moment value is poorer.
the present invention has beneficial effect.the present invention adopts same quefrency-inverse singular frequency method, obtain the self-adaptation combined invariant moment value with weight properties, reduce not bending moment value, compose with different weights to the not bending moment value of different contribution, carry it in multi classifier and calculate, effectively can reduce the time of calculating, promote the performance identified, obtain good classification results.Self-adaptation combined invariant moment value method is adopted all to have best discrimination in different scene, as can be seen here, different not bending moments has different contribution functions to different scenes, choose suitable not bending moment value and assign weight to not bending moment value and contribute to adapting to different scene, obtain higher discrimination.Self-adaptation combined invariant moment has simplified part not bending moment value by the calculating in early stage, more particularly expends the high-order not bending moment value of more time, so its computing time is moderate, is adapted to completely in the real-time monitoring of video monitoring.
Accompanying drawing explanation
Fig. 1 is similar frequency computation part schematic diagram;
Fig. 2 is the flow process of the characteristics of objects expression of doing more physical exercises being applicable to different scene;
Fig. 3 is the whole equal weight of road monitoring scene not bending moment value curved surface schematic diagram;
Fig. 4 is road monitoring scene adaptive not bending moment value weight curved surface schematic diagram;
Fig. 5 is the whole equal weight of river course monitoring scene not bending moment value curved surface schematic diagram;
Fig. 6 is river course monitoring scene self-adaptation not bending moment value weight curved surface schematic diagram;
Fig. 7 is the whole equal weight of cell monitoring scene not bending moment value curved surface schematic diagram;
Fig. 8 is cell monitoring scene adaptive not bending moment value weight curved surface schematic diagram;
Fig. 9 is various combination not bending moment value discrimination comparison diagram.
Embodiment
Provide the example that the present invention is directed to different motion object implementation process in different scene below, outdoor scene comprises: road Road, river course River, community ResidentialArea tri-kinds of situations.
embodiment 1
Example in road monitoring scene, flow process as shown in Figure 2.
The first step: choose { general car Car, pedestrian Person, bus Bus, middle bus Van, bicycle Bicycle} Moving Objects classification.
Orientate the N number of Moving Objects occurred in traffic surveillance videos as class Class, class set is combined into { CCar, CPerson, Cbus, Cvan, CBicycle}, then in video to the concrete object of such correspondence of the corresponding extraction of each class, CCar class correspondence extracts { Car1, Car2, Car3, Cari}, CPerson class correspondence extracts { Person1, Person2, Person3, Personi}, CBus class correspondence extracts { Bus1, Bus2, Bus3, Busi}, CVan class correspondence extracts { Van1, Van2, Van3, Vani}, CBicycle class correspondence extracts { Bicycle1, Bicycle2, Bicycle3, Bicyclei}.All kinds of not bending moment values each Car object being calculated respectively to its various angle form as:
,
,
, the computation process of the Moving Objects reference Car object in other road monitorings,
Wherein, M for initially to adopt not bending moment value dimension, N be for such all instance objects numbers, Q is instance objects various angle form number, m≤M, i≤N, j≤Q.
By the mode of averaging
unified various angle, then calculated by formula (1), obtain original input data as follows:
,
,
,
, the like, calculate all primary datas.
Second step: { in CCar, CPerson, Cbus, Cvan, CBicycle}, the not bending moment value of each Moving Objects carries out to Moving Objects set
the calculating of value;
According to frequency calculation method, the C1 value of Moving Objects Car is mapped on linear coordinate, calculates the average of all not bending moment values
, then not bending moment value frequency be
find in scope other
number, by parity of reasoning, calculates Moving Objects all
value.
According to similar frequency calculation method, all kinds Car's
bending moment does not calculate Data distribution8 as shown in Figure 1.Then for Car1's
value scope can match Car2's
value, therefore Car1
the numerical value of SF be 1, can calculate Car2's thus
sF value be 2, Car3's
sF value be 1, Car4's
sF value be 0.
According to inverse singular frequency computing method, if certain experimental data comprises 100 sample datas, not bending moment value
occur 10 times in different Moving Objects coupling, another is bending moment value not
occur 1 time in the coupling of different motion object, so
ratio
there is better discrimination.
Formula (2) is adopted to calculate all not bending moment values of each Moving Objects in road monitoring
's
value, utilizes these
value sets up a vector model for each Moving Objects, and it can represent the criteria for classification for feature, the related check standard of each dimension, and in road monitoring scene, the vector model of Car is:
In road monitoring scene, the vector model of Person is:
In road monitoring scene, the vector model of Bus is:
In road monitoring scene, the vector model of Van is:
In road monitoring scene, the vector model of Bicycle is:
。
3rd step: not bending moment value in Moving Objects all in road monitoring scene
value sorts, and therefrom chooses
value is greater than superseded threshold value
the not bending moment value of=0.6 as key not bending moment value, using this key not bending moment value as the character representation of Moving Objects, to original all not bending moment value carry out dimensionality reduction operation, the cancellation poor efficiency not bending moment value of 66%, so efficiency is enhanced.
4th step: because the characteristic type of each Moving Objects is not quite similar, thus the dimension characterizing the not bending moment value value weight of each scene is also different, these impacts must be eliminated, make the not bending moment value value weight of scene class meet the characteristic type of all kinds of Moving Objects.
If
,
the not bending moment value weight distribution of Moving Objects CCar and CBus,
,
, wherein M is all not bending moment value numbers, m≤M;
If the not bending moment value that in Moving Objects CCar and CBus, similarity is higher is each other more, and not shared by bending moment value
value ratio in respective Moving Objects is higher, illustrate these not bending moment value more can reflect their importance in Moving Objects, therefore according to the key that meets similarity threshold condition in key not bending moment value not bending moment value
value is at whole Moving Objects
ratio shared in value summation carries out weight calculation; Calculated by formula (4)
etc. all Similarity value, calculated further by formula (3)
value; According to formula (5), if key not bending moment value
in certain key not bending moment value weight
with another key not bending moment value
in key not bending moment value weight
similarity exceed user setting similarity threshold
=0.7, then by this crucial bending moment value weight
put into set
, calculate all crucial bending moment value weight similarity relations successively.
5th step: calculate all combined invariant moment value weights in this scene, because bending moment value does not represent most important character representation in a Moving Objects, therefore the not bending moment value weight of road monitoring scene just can by Moving Objects { CCar, CPerson, Cbus, the not bending moment value weight of Cvan, CBicycle} comes out, and calculates this scene successively choose all not bending moment values by formula (6)
.
6th step: by Weight
combined invariant moment value
put into the training of Support Vector Machines for Regression model, kernel function adopts radial basis function, Support Vector Machines for Regression training pattern can be obtained by training study, using combined invariant moment value in test set as input parameter, calculate in Support Vector Machines for Regression model after input training, object classification result of doing more physical exercises can be obtained.
As shown in Figure 3, in road monitoring scene, the not bending moment weight fold curved surface of Moving Objects Car presents big bump state, reflect that the contribution of not bending moment value to this scene motion Object identifying of this scene has a long way to go with this, by method in the present invention to after the reselecting of not bending moment value as shown in Figure 4, in road monitoring scene, the not bending moment weight fold curved surface of Moving Objects Car is relatively steady, illustrate after bending moment value does not take adaptive measure of sifting out, the individual difference of different not bending moment value can be reduced, select representative not bending moment value to combine, reach good classifying quality.
embodiment 2
Example in river course monitoring scene:
The first step: choose { canoe Boat, general car Car, little crane SmallCrane, medium-sized crane Medium-sizedCrane, pedestrian Person} Moving Objects classification.
Orientate the N number of Moving Objects occurred in river course monitor video as class Class, class set is combined into { CBoat, CCar, CSmallCrane, CMedium-sizedCrane, CPerson}, then in video to the concrete object of such correspondence of the corresponding extraction of each class, CBoat class correspondence extracts { Boat1, Boat2, Boat3 ... Boati}, the like, calculate the concrete object of all categories.
All kinds of not bending moment values each Boat object being calculated respectively to its various angle form as:
,
,
employing average and formula (1) to calculate original input data further as follows:
,
,
,…
Moving Objects in the monitoring of other river courses is with reference to the computation process of Boat object.
Second step: adopt formula (2) to calculate all not bending moment values of each Moving Objects in river course monitoring
's
value, utilizes these
value sets up a vector model for each Moving Objects, and it can represent the criteria for classification for feature, the related check standard of each dimension, and in river course monitoring scene, the vector model of Boat is:
The rest may be inferred for other vector model generation methods.
3rd step: not bending moment value in Moving Objects all in river course monitoring scene
value sorts, and therefrom chooses
value is greater than superseded threshold value
the not bending moment value of=0.8 is as key not bending moment value, using this key not bending moment value as the character representation of Moving Objects, to original all not bending moment value carry out dimensionality reduction operation, the cancellation poor efficiency not bending moment square value of 81%, so efficiency is enhanced.
4th step: establish
,
the not bending moment value weight distribution of Moving Objects CBoat and CSmallCrane,
,
, wherein M is all not bending moment value numbers, m≤M; Calculated by formula (4)
etc. all Similarity value, and calculated further by formula (3)
value; According to formula (5), if key not bending moment value
in certain key not bending moment value weight
with another key not bending moment value
in key not bending moment value weight
similarity exceed user setting similarity threshold
=0.5, then by this crucial bending moment value weight
put into set
, calculate all crucial bending moment value weight similarity relations successively.
5th step: calculate this scene successively by formula (6) and choose all not bending moment values
.
6th step: bring Support Vector Machines for Regression model into by generating data above, generates the Support Vector Machines for Regression model after training and is used for classification.
As shown in Figure 5, in river course monitoring scene, the not bending moment weight fold curved surface of Moving Objects Boat presents big bump state, reflect that the contribution of not bending moment value to this scene motion Object identifying of this scene has a long way to go with this, by method in the present invention to after the reselecting of not bending moment value as illustrated by fig. 6, in river course monitoring scene, the not bending moment weight fold curved surface of Moving Objects Boat is relatively steady, illustrate after bending moment value does not take adaptive measure of sifting out, the individual difference of different not bending moment value can be reduced, select representative not bending moment value to combine, reach good classifying quality.
embodiment 3
Example in cell monitoring scene
The first step: choose { bicycle Bicycle, in-between car Van, general car Car, pedestrian Person} Moving Objects classification.
Orientate the N number of Moving Objects occurred in cell monitoring video as class (Class), class set is combined into { CBicycle, CVan, CCar, CPerson}, then in video to the concrete object of such correspondence of the corresponding extraction of each class, CBicycle class correspondence extracts { Bicycle1, Bicycle2, Bicycle3, Bicyclei}, the like, calculate the concrete object of all categories.
All kinds of not bending moment values each Bicycle object being calculated respectively to its various angle form as:
,
,
Employing average and formula (1) to calculate original input data further as follows:
,
,
,…
Moving Objects in other cell monitorings is with reference to the computation process of Bicycle object.
Second step: all not bending moment values adopting each Moving Objects in the monitoring of formula (2) calculation plot
's
value, utilizes these
value sets up a vector model for each Moving Objects, and it can represent the criteria for classification for feature, the related check standard of each dimension, and in cell monitoring scene, the vector model of Bicycle is:
The rest may be inferred for other vector model generation methods.
3rd step: not bending moment value in Moving Objects all in cell monitoring scene
value sorts, and therefrom chooses
value is greater than
the not bending moment value of=0.7 is as key not bending moment value, using this key not bending moment value as the character representation of Moving Objects, to original all not bending moment value carry out dimensionality reduction operation, the cancellation poor efficiency not bending moment square value of 75%, so efficiency is enhanced.
4th step: establish
,
the not bending moment value weight distribution of Moving Objects CBicycle and CVan,
,
, wherein M is all not bending moment value numbers, m≤M; Calculated by formula (4)
etc. all Similarity value, and calculated further by formula (3)
value; According to formula (5), if key not bending moment value
in certain key not bending moment value weight
with another key not bending moment value
in key not bending moment value weight
similarity exceed user setting similarity threshold
=0.6, then by this crucial bending moment value weight
put into set
, calculate all crucial bending moment value weight similarity relations successively.
5th step: calculate this scene successively by formula (6) and choose all not bending moment values
.
6th step: bring Support Vector Machines for Regression model into by generating data above, generates the Support Vector Machines for Regression model after training and is used for classification.
As shown in Figure 7, in cell monitoring scene, the not bending moment weight fold curved surface of Moving Objects Person presents big bump state, reflect that the contribution of not bending moment value to this scene motion Object identifying of this scene has a long way to go with this, by method in the present invention to after the reselecting of not bending moment value as illustrated by fig.8, in cell monitoring scene, the not bending moment weight fold curved surface of Moving Objects Person is relatively steady, illustrate after bending moment value does not take adaptive measure of sifting out, the individual difference of different not bending moment value can be reduced, select representative not bending moment value to combine, reach good classifying quality.
As shown in Figure 9, for the Hu calculated separately not bending moment, polar radius not bending moment, affine not bending moment, Wavelet Invariant Moment and combination fixing Hu not bending moment add affine not bending moment and adaptive H u not bending moment add affine not bending moment, self-adaptation combined invariant moment value discrimination is the highest.
Claims (4)
1. be applicable to a characteristics of objects expression of doing more physical exercises for different scene, it is characterized in that comprising the following steps:
The first step: by all kinds of initial not bending moment value C' of object of doing more physical exercises in scene
mi, C'
mirepresent i-th of m object initial not bending moment value, taking absolute value according to formula (1) to not bending moment and take the logarithm calculates not bending moment value C
mi, C
mirepresent i-th of m object not bending moment value
C
mi=lg|C'
mi|(1)
Second step: the not bending moment value C calculating each Moving Objects in whole Moving Objects set
misame quefrency-inverse singular frequency SF-ISF value
SF-ISF(C
mi)=sf(C
mi)×log(N/isf(C
mi))(2)
Sf (C in formula
mi) represent not bending moment value C
mithe frequency occurred in same Moving Objects; N represents the sum of all Moving Objects in Moving Objects set; Isf (C
mi) represent the frequency occurring current not bending moment value in different motion object set, by each in Moving Objects set not bending moment value carry out above-mentioned analysis, obtain the SF-ISF value of each not bending moment value of each Moving Objects; And then utilize SF-ISF value to set up vector model respectively for each Moving Objects, as the criteria for classification of feature and the related check standard of each dimension;
3rd step: according to the vector model described in step 2, sorts to the SF-ISF value of not bending moment value, therefrom chooses the SF-ISF value of the superseded threshold value P being greater than user's setting; Select the key not bending moment value C that SF-ISF value is corresponding
mj, C
mjrepresent a jth key not bending moment value of m object, 1≤j≤J, J is the number of all keys not bending moment value, by C
mjas the feature of Moving Objects;
4th step: a jth key not bending moment value C calculating m Moving Objects of described step one Scene
mjweight sets W
mj, W
mj={ w
1j, w
2j... w
mj, w
i,jcomputing method as shown in formula (3), 1≤i≤m
The computing method of described avg (i, j) are as shown in formula (4)
SF-ISF (C in formula (4)
ik) represent key not bending moment value C
iksF-ISF value;
∧
ithe key not bending moment value weight w of the similarity threshold μ of user's setting is exceeded for similarity
jlset
5th step: the not bending moment vector weighted value wf calculating described step one Scene
M represents the sum of the Moving Objects having chosen not bending moment value in Moving Objects set;
Calculate this scene according to formula (6) and choose all not bending moment value { wf
1, wf
2... wf
n}
6th step: by Weight { wf
1, wf
2... wf
ncombined invariant moment value { C
1, C
2... C
nput into the training of multi classifier model, obtain training pattern by training study, using the combined invariant moment value in practical application as input parameter, calculate in the model after input training, can obtain doing more physical exercises in scene object classification result.
2. a kind of characteristics of objects expression of doing more physical exercises being applicable to different scene as claimed in claim 1, is characterized in that described frequency calculation method is as follows:
C
mrepresent the not bending moment of m object, C
mirepresent i-th of m object not bending moment value, X represents the number of all not bending moment values, by the C of all Moving Objects
mivalue is mapped on linear coordinate, calculates the average of all not bending moment values
not bending moment value C
mifrequency be
the number of other the not bending moment values than itself in scope.
3. a kind of characteristics of objects expression of doing more physical exercises being applicable to different scene as claimed in claim 1, is characterized in that described same quefrency is certain numerical value that bending moment value is not adjacent in same Moving Objects; Adjacent numerical value is more, and bending moment value is more not relevant with the character representation of this Moving Objects to represent this.
4. a kind of characteristics of objects expression of doing more physical exercises being applicable to different scene as claimed in claim 1, is characterized in that described inverse singular frequency is certain numerical value that bending moment value is not adjacent in different Moving Objects coupling; Adjacent numerical value is more, represents that the separating capacity of this not bending moment value is poorer.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310232933.9A CN103279769B (en) | 2013-06-13 | 2013-06-13 | A kind of characteristics of objects expression of doing more physical exercises being applicable to different scene |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201310232933.9A CN103279769B (en) | 2013-06-13 | 2013-06-13 | A kind of characteristics of objects expression of doing more physical exercises being applicable to different scene |
Publications (2)
Publication Number | Publication Date |
---|---|
CN103279769A CN103279769A (en) | 2013-09-04 |
CN103279769B true CN103279769B (en) | 2016-02-24 |
Family
ID=49062284
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201310232933.9A Expired - Fee Related CN103279769B (en) | 2013-06-13 | 2013-06-13 | A kind of characteristics of objects expression of doing more physical exercises being applicable to different scene |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103279769B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103278570A (en) * | 2013-06-13 | 2013-09-04 | 江苏大学 | Ultrasonic linear phased array transducer for detecting metallic material and manufacturing method |
CN113140107B (en) * | 2020-03-30 | 2022-09-16 | 北京高德云图科技有限公司 | Traffic information processing method and device, electronic equipment and computer storage medium |
CN115138547A (en) * | 2022-06-30 | 2022-10-04 | 中国工程物理研究院电子工程研究所 | Piezoelectric micromechanical ultrasonic transducer back lining material, preparation method and filling method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101464948A (en) * | 2009-01-14 | 2009-06-24 | 北京航空航天大学 | Object identification method for affine constant moment based on key point |
CN101847258A (en) * | 2009-03-26 | 2010-09-29 | 陈贤巧 | Optical remote sensing image registration method |
CN102324045A (en) * | 2011-09-01 | 2012-01-18 | 西安电子科技大学 | Invariant-moment target recognition method based on Radon transformation and polar harmonic transformation |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7840059B2 (en) * | 2006-09-21 | 2010-11-23 | Microsoft Corporation | Object recognition using textons and shape filters |
-
2013
- 2013-06-13 CN CN201310232933.9A patent/CN103279769B/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101464948A (en) * | 2009-01-14 | 2009-06-24 | 北京航空航天大学 | Object identification method for affine constant moment based on key point |
CN101847258A (en) * | 2009-03-26 | 2010-09-29 | 陈贤巧 | Optical remote sensing image registration method |
CN102324045A (en) * | 2011-09-01 | 2012-01-18 | 西安电子科技大学 | Invariant-moment target recognition method based on Radon transformation and polar harmonic transformation |
Also Published As
Publication number | Publication date |
---|---|
CN103279769A (en) | 2013-09-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108510532B (en) | Optical and SAR image registration method based on deep convolution GAN | |
CN109359559B (en) | Pedestrian re-identification method based on dynamic shielding sample | |
Tao et al. | Smoke detection based on deep convolutional neural networks | |
CN107563372B (en) | License plate positioning method based on deep learning SSD frame | |
Huttunen et al. | Car type recognition with deep neural networks | |
CN109215034B (en) | Weak supervision image semantic segmentation method based on spatial pyramid covering pooling | |
CN103177458B (en) | A kind of visible remote sensing image region of interest area detecting method based on frequency-domain analysis | |
Chen et al. | Research on recognition of fly species based on improved RetinaNet and CBAM | |
CN104537647A (en) | Target detection method and device | |
CN106960176B (en) | Pedestrian gender identification method based on transfinite learning machine and color feature fusion | |
CN104268593A (en) | Multiple-sparse-representation face recognition method for solving small sample size problem | |
Tang et al. | Deep fishernet for object classification | |
CN106529570B (en) | Image classification method based on depth ridge ripple neural network | |
CN105787501A (en) | Vegetation classification method capable of automatically selecting features in power transmission line corridor area | |
Guo et al. | Network pruning for remote sensing images classification based on interpretable CNNs | |
CN114255403A (en) | Optical remote sensing image data processing method and system based on deep learning | |
CN110443155A (en) | A kind of visual aid identification and classification method based on convolutional neural networks | |
CN103279769B (en) | A kind of characteristics of objects expression of doing more physical exercises being applicable to different scene | |
Yang et al. | Cross modality knowledge distillation for multi-modal aerial view object classification | |
CN116152678A (en) | Marine disaster-bearing body identification method based on twin neural network under small sample condition | |
Li et al. | Slim-neck by GSConv: a lightweight-design for real-time detector architectures | |
Yin et al. | An enhanced lightweight convolutional neural network for ship detection in maritime surveillance system | |
CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
Xiong et al. | Oil spills identification in SAR image based on convolutional neural network | |
Zhang et al. | Wood board image processing based on dual-tree complex wavelet feature selection and compressed sensing |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160224 Termination date: 20190613 |
|
CF01 | Termination of patent right due to non-payment of annual fee |